The present invention generally relates to systems for integrated circuits. More particularly, the present invention relates to a power delivery system, a signal transfer system, a package design system, a thermal management system, and an electromagnetic interference (EMI) emission control system for an integrated circuit to support advancements in semiconductor technology.
I. Semiconductor Technology
Consumers demand innovative electronic products that have more functionality, better performance, smaller size, less weight, better reliability, lower cost and faster time-to-market. Semiconductor technology is the core building block for the innovative electronic products desired by consumers. Over the years, advancements in semiconductor technology have led to dramatic increases in the functionality and performance of integrated circuit (IC) devices while minimizing the size, weight, defects and cost of the IC devices.
Historically, the number of transistors that the electronic industry can place on a semiconductor chip doubles about every eighteen months. This rapid development cycle permits fast delivery of the new innovative products to the market. For example, semiconductor manufacturers took nearly thirty years to perfect microprocessor clock rates to run at 1 GHz, yet manufacturers recently reached the 2 GHz microprocessor clock rate less than eighteen months after reaching 1 GHz. Manufacturers anticipate that there are no fundamental barriers to extending the rapid advancement of semiconductor technology for another decade by building the even faster silicon transistors. These transistors are anticipated to be around 20 nanometers (nm) in size and should permit the manufacturers to build microprocessors containing a billion transistors which run at speeds approaching 20 GHz and operate at less than one volt within the next few years. These new transistors, which act like switches controlling the flow of electrons inside a microprocessor, will turn on and off more than a trillion times per second. Such advancements in semiconductor technology will result in microprocessors that have faster clock rates, higher power, lower supply voltages, higher DC currents, higher transient currents, narrower voltage margins, higher non-uniform heat densities, and higher frequency electromagnetic interference emissions. Ancillary benefits to these advancements include microprocessors that have increased interconnect densities, reduced circuit board real estate and package volume, and improved product manufacturing and reliability.
Specifications for near future microprocessors require 1.0V operating voltage, 100 A current, 300 A/μsec transient currents, efficiency greater than 90%, regulation within 5%, and voltage ripple less than 1%. These requirements present a significant advancement over present microprocessor designs. Microprocessors having these characteristics and requirements and future microprocessors having even more demanding characteristics and requirements will need new support systems, such as power delivery, signal transfer, packaging, thermal management, and electromagnetic interference (EMI) emission control.
II. Power Delivery
Power delivery concerns supplying power to devices that need it. Traditionally, an ideal power supply is assumed and little consideration is given to power delivery until the end of the design. Printed circuit board (PCB) designers attempt to create the ideal power delivery supply with conventional power and ground planes in the PCB and with wide, heavy traces on the PCB to distribute the power among the devices on the PCB. High frequency ceramic capacitors control high frequency noise, created by switching the transistors on and off, by shorting the high frequency noise to ground. Lower frequency bulk capacitors (such as tantalum capacitors) subsequently recharged the high frequency ceramic capacitors. Various rules of thumb exist for determining the amount of each type of capacitance that is required for various ICs.
To electrically model this power delivery system, considerations include the inductance and resistance of cables, connectors, PCB, pins, contacts and components, such as resistors and capacitors, of the receiving device(s) and power source(s). In the past, voltage drops due to inductance (V=L di/dt) and resistance (V=IR) have been nearly negligible relative to the tolerance of devices in most systems. Similarly, simple rules of thumb determine the method for decoupling the high frequency noise.
Each generation of semiconductor technology has reduced power supply voltage to support the requirements of deep sub-micron semiconductor technologies and to improve reliability. Lower power supply voltages should lower the power consumption. However, even at lower power supply voltages the power consumption of microprocessors is increasing because of more transistors, increased density of transistors on the die, thinner insulators that increase capacitance, and higher operating frequencies. Power consumption in microprocessors continues to rise as much as three times every two years while microprocessor power supply voltages approach 1.0 V. Power consumption (P) is related to the operating frequency (f), the power supply voltage (V), and the chip capacitance (C) of the microprocessor by the formula (P=CfV2). By one example, a microprocessor with a typical chip capacitance of 20 nanofarads, a power supply voltage of 1.65 volts, and an operating frequency of 1 GHz, will consume 55 watts of power (0.020×1.65×1.65×1,000). By another example, a microprocessor with a typical chip capacitance of 40 nanofarads, a power supply voltage of 1 volt, and an operating frequency of 3 GHz, will consume 120 watts (0.040×1.0×1.0×3,000).
Power consumption (P) is also related to the power supply voltage (V) and the current (I) by the formula (P=VI). This formula shows that high power consumption (P) at low power supply voltages (V) requires that high currents (I) (I=P/V) be delivered to the microprocessor. Continuing with the two examples above, the microprocessor consuming 55 watts of power and having a power supply voltage of 1.65 volts requires a supply current of 33 amps (55/1.65), and the microprocessor consuming a 120 watts of power and having a power supply voltage of 1.0 volts requires a supply current of 120 amps (120/1), representing an increase of about 3.6 times over the 33 amp microprocessor.
At these voltage and current levels, it is more difficult for a central power supply to deliver high current and low voltage power throughout a computer system because of impedance levels that cause unacceptable voltage drops along the power distribution paths. Computer systems presently use distributed power systems to route power throughout the computer system at high voltage and low current and then convert to low voltage and high current as needed by the microprocessor. Voltage regulators or modular DC/DC converters, which provide the needed low voltage, high current power, are located as close as possible on the motherboard to the microprocessor to minimize the impedances and the resulting voltage drops. The location of the power distribution path on the mother board takes up valuable space that could be used for other components.
Even with distributed power delivery systems, every part of the distribution path must still have a low impedance to minimize the resulting voltage drops. Typically, the voltage variance at the voltage regulator is less than (e.g., about one-half) of the voltage variance at the microprocessor. Traditionally, connectors with a high pin count and heavy copper power/ground planes are used to minimize the impedance. However, these solutions also consume extra printed circuit board space and add cost.
In one power distribution approach, the microprocessor and the voltage regulator each form modules and rely upon corresponding sockets to connect each module to the motherboard. The microprocessor may be mounted to an interposer board, and the motherboard has one socket that receives the voltage regulator and another socket that receives the interposer board. The microprocessor and voltage regulator are modular for fast and easy exchange for efficient manufacturing and service. Current flows from the voltage regulator to the microprocessor over a path from the voltage regulator, through its socket, the motherboard, the interposer socket and board, the microprocessor package, and ends at the die. This relatively long path of current flow introduces impedance and voltage drops, which are not desirable for advanced microprocessor designs.
An alternative power system approach bypasses the motherboard and the microprocessor socket. In this approach, the interposer board carries the microprocessor die and the voltage regulator. Current flows from the voltage regulator to the microprocessor over a path starting from the voltage regulator, through the voltage regulator socket, the interposer board, the microprocessor package, and ending with the die. Since this approach bypasses the motherboard and the interposer socket, the path of current flow is shorter. Therefore, this approach improves the impedance and the resulting voltage drop of the relatively shorter path.
Someday it may be possible to integrate the voltage regulator into the microprocessor package, making the path of current flow very short, reducing the impedance and resulting voltage drop. However, semiconductor technology has not advanced far enough to provide this level of an integrated system.
Microprocessor response time or transient current requirement (di/dt), i.e., the rate at which the current demand changes is another power-related concern. Varying computing demands of the microprocessor requires varying current demands from the power supply. The computing demands vary because of high clock speed circuits and power conservation design techniques, such as clock gating and sleep modes. These techniques result in fast, unpredictable and large magnitude changes in supply current ultimately requiring hundreds of amps within a few nanoseconds. The resulting current surge demanded by the microprocessor from the voltage regulator can cause unacceptable voltage spikes on the power delivery voltage according to the formula (dV=IR+Ldi/dt).
Attempts have been made to manage surge currents by placing decoupling capacitors throughout the power delivery system such as on the voltage regulation module, the motherboard, the interposer PCB, the die package, and on the die itself. Decoupling capacitors are typically located on the circuit board outside the microprocessor package, typically using several discrete decoupling capacitors mounted next to the microprocessor package on the circuit board. In this approach, conductive traces on the circuit board connect the decoupling capacitors to power and ground pins on the microprocessor. In another approach, a discrete decoupling capacitor is formed as part of the IC.
These decoupling capacitors are commonly used to ensure that the power supply system can provide the microprocessor with a surge current when required. The decoupling capacitors connect power sources to the power leads of the microprocessor. The amount of decoupling capacitance needed depends on the power requirement of the microprocessor. The microprocessor is able to draw its required surge current from the power stored in the decoupling capacitors, and hence, the decoupling capacitors stabilize the power delivery system by storing power local to the microprocessor in order to meet the surge current needs of the microprocessor. However, use of discrete, broad-mounted decoupling capacitors not only increase the cost of the power delivery system, but also consume additional area on the IC or the circuit board, or elsewhere.
As the power requirement of microprocessor increases, the need for more decoupling capacitance increases, which in turn requires larger value or size decoupling capacitors and more space to accommodate them. Unfortunately, larger value or size decoupling capacitors consume more area on the circuit board.
As the switching speeds of the transistors increases, an undesirable amount of resistance due to inductance, associated with the interconnection between the semiconductor die and the decoupling capacitor, increases according to the formula (XL=2fL). The longer the conductive path interconnecting the decoupling capacitor and the semiconductor die inside the microprocessor, the higher the inductance. The higher the frequency of operation of the microprocessor, the higher the resistance of the system due to the inductance, and higher resistance causes a higher voltage drop. Therefore, it is desirable to locate the decoupling capacitors as close to the semiconductor die as possible, such as by putting the decoupling capacitor inside the microprocessor package, as described above, in order to minimize the conductive path to minimize the inductance.
Further, capacitors exhibit inductance and resistance characteristics as well as capacitance characteristics and can be electrically modeled as a series RLC circuit. At higher frequencies, such as above 100 MHz, the inductance characteristic limits the effectiveness of conventional discrete decoupling capacitors. If large surge currents are required by the microprocessor, this residual inductance can cause unacceptable voltage drops and AC noise.
Historically, power has been brought to the IC through pins in the IC socket. As the power requirements of an IC increase, it will require additional pins to accommodate the power, and these additional pins increase the size of the IC package and therefore take up valuable space on the circuit board. The increase in the pin numbers also increases the amount of force required for inserting the IC into and removing it from its socket of the circuit board. The power pins are run through the same surface of the IC, typically the bottom surface, and with high densities, the power and signal pins should be isolated from each other to prevent crosstalk and noise.
Hence, there is a need for a power delivery system that delivers low voltage, narrow voltage margin, high current, and high transient current to a high performance integrated circuit, such as a microprocessor, that minimizes cost and space while improving reliability.
III. Signal Transfer
Signal integrity is a complex field of study involving digital and analog design, circuit, and transmission line theory and involves phenomenon such as cross talk, ground bounce, and power supply noise. Although signal integrity has always been important, in the past the switching speed of microprocessor transistors was so slow that digital signals actually resembled high pulses, representing ones, and low pulses, representing zeros. Electrical modeling of signal propagation was often not necessary. Unfortunately, at today's microprocessor speeds of 1 GHz and above even the simple, passive elements of a high-speed design, such as wires, PC boards, connectors, and microprocessor packages, can significantly affect the wave shape and voltage level of the signal. Further, these passive elements can cause glitches, resets, logic errors, and other problems.
Typically, a microprocessor makes contact with the motherboard using galvanic (i.e., metal-to-metal) connections such as a land grid array (LGA), ball grid array (BGA), pin grid array (PGA) and solder, to transfer signals between the microprocessor and the motherboard. As the switching speeds of the transistors increases, an undesirable amount of resistance due to inductance, associated with the conductive interconnection between the semiconductor die located inside the microprocessor and the motherboard, increases according to the formula (XL=2fL). The longer the conductive path interconnecting the semiconductor die in the microprocessor to the motherboard, the higher the inductance. A higher frequency of operation of the microprocessor causes a higher resistance due to the inductance on the signal path, and this resistance causes a higher voltage drop of the signal level. Therefore, it is desirable to minimize the inductance of the signal path as the frequency of operation of the microprocessor increases. Other disadvantages of signaling via conductive contacts are disclosed in U.S. Pat. No. 5,629,838, issued May 13, 1997. An engineering tradeoff exists between increasing the desired operating frequency of the microprocessor and the signal integrity of the system.
Hence, there is a need for a system that permits the operating frequency of the microprocessor to increase without degrading the integrity of the signal. Such a system would maximize the performance and minimize the cost of interconnection technology used in high-speed digital signal designs.
IV. Integrated Circuit package Design
Advances in semiconductor technology provide microprocessors that have higher performance and are smaller in size, which directly affects the design of the microprocessor package. Factors related to microprocessor package design include: current per contact and per socket, the number of ground and power pins, the number of signal contacts and signal contacts per square area, the contact pitch, the number of total contacts and total contacts per square area, the contact force along the Z-axis, the mated contact height, the signal bandwidth, the semiconductor die size, and other factors.
Increasing the number and power of transistors in the microprocessor typically increases current per contact and socket as well as increases the number of ground and power pins. Increasing the performance of the microprocessor will need an increase in the number of signal contacts and the semiconductor die size. Increasing both the power and performance of the microprocessor will increase the total contacts and will decrease the contact pitch. Increasing the number of total contacts while decreasing the contact pitch will increase the contact force required along the Z-axis which may require an increase in the mated contact height. Increasing the frequency of operation of the microprocessor will decrease the signal bandwidth. Hence, it should be understood that engineering tradeoffs exist among these factors in order to produce a microprocessor having an optimized package design.
V. Thermal Management
Advances in electronic packaging design provide devices with higher performance and smaller size, which lead to increased heat generation and heat density, which in turn may cause thermal management to be given higher priority in package design to maintain reliability of the device.
For microprocessors, higher performance, increased level of integration, and optimization of die size has led to higher non-uniform heat density in certain areas of the microprocessor die. Heat generation and heat density continue to increase with more advanced semiconductor technology. The reliability of a microprocessor is exponentially dependent on the operating temperature of the die junction, which depends on the power consumed by the transistor having the die junction.
Thermal management of the microprocessor is related to thermal management of the voltage regulator. Both the efficiency of the voltage regulator and the power consumed by the processor must be considered together. For example, a voltage regulator operating at 85 percent efficiency and which drives a microprocessor consuming 120 watts of power, dissipates about 18 watts of power. This power must be drawn away from the voltage regulator and microprocessor to cool the devices in order to maintain their reliability. Therefore, an engineering tradeoff exists between locating the voltage regulator near the microprocessor to minimize impedance and the resulting voltage drop, as described above, and locating the voltage regulator far from the microprocessor to minimize the heat generation and heat density.
Hence, there is a need for a thermal management solution that permits a high power microprocessor to be located near voltage regulator to minimize the impedance and resulting voltage drop while efficiently dissipating heat generation and heat density to maximize reliability.
VI. Electromagnetic Interference
Sources of electromagnetic interference (EMI) emission include the transistors within a microprocessor and signal paths on circuit boards and cables. The microprocessor is one of the largest sources of EMI in computer systems. Microprocessor clock signals have increased in frequency to 1 GHz and beyond today. At 1 GHz, these clock signals can generate harmonic frequency signals that reach 5 Ghz, and both of these signals generate EMI waves with wavelengths that are inversely proportional to the frequency of the signal (i.e., the higher the frequency, the shorter the wavelength).
Typically, a conductive shield or cover is used to control EMI. The shield is grounded to provide a dissipating path for the EMI to prevent it from interfering with other circuits. The shield usually contains holes for thermal management to create airflow to cool the device generating the EMI. However, large holes in the shield permit EMI to escape through the shield, and thus the shield holes must be sized so that the EMI does not escape, but airflow is not restricted to cool the device. High frequency signals require smaller holes in the shield for EMI containment, but the smaller holes restrict the airflow available for cooling. Hence, an engineering tradeoff exists in sizing the holes in the shield to for cooling and EMI containment purposes.
The shield may be located at the microprocessor or chassis level, or both. The microprocessor generates the high frequency harmonic signals that cause EMI, so that locating the shield close to the microprocessor may effectively contain the harmonic signals near the source of the EMI. Localized containment prevents the EMI from interfering with other circuitry in the computer system, but it also restricts the airflow needed to dissipate the microprocessor heat. Alternatively, the chassis of the computer system may be used as the shield which improves the airflow around the microprocessor but permits EMI to interfere with other circuits in the system. A chassis level solution requires small holes in the chassis for EMI blockage, but reduces airflow.
Grounding a heat sink that located near the microprocessor is another way to reduce EMI. However, EMI from the microprocessor that couples with the heat sink may cause the heat sink to act as an antenna and radiate the EMI. It is difficult to ground the heat sink through the microprocessor package, and although grounding the heat sink may reduce EMI, this solution alone may not be sufficient to pass required FCC emission tests. Additional shielding may be necessary to block the EMI. Therefore, there is a need for an EMI containment system that contains EMI from high frequency signals without compromising the thermal management of the system.
In summary, systems related to power delivery, signal transfer, package design, thermal management, and electromagnetic interference (EMI) emission control for an integrated circuit are needed to support future and current advancements in semiconductor technology.